1 author: Mira Ariel Tel Aviv University 59

Grammar proposes discourse disposes

Discourse, as we have noted above, is not a random collection of sentences. Speakers have certain communicative goals when they engage in discourse. These goals often require a lot of inferencing above the linguistic decoding of linguistic strings. One thing speakers can do for addressees is choose their linguistic utterances in such a way that will facilitate the addressees in their
Ariel: Discourse, grammar, discourse 9
interpretations. This is why the constructions we actually find in discourse do not reflect a random selection of conceptually appropriate and grammatically well-formed strings. For any kind of message we wish to convey (content-wise), more often than not, grammar will make available more than one form. We have semantic paraphrases (e.g. I must get some soda – LSAC versus ~I have to get some soda), syntactic paraphrases (e.g. Chelsea was booed off by their own fans
– WebCorp, versus ~Their own fans booed Chelsea off) and phonological para- phrases, or variants (e.g. the horse kinda helps you out – SBC: 001, versus ~ the horse kind of helps you out).³ Such grammatical and conceptual equivalents are not necessarily discoursally equivalent, however. Some variants better serve the speaker’s current discourse needs or goals than others. Section 2 presents three examples for how, when and why speakers strategically select certain but not other grammatical forms in specific discourse contexts. They exemplify what I mean by ‘grammar proposing, discourse disposing’.

1. 
   THE I THINK DISCOURSE PATTERN
   

The first example concerns the grammatically optional occurrence of com- plementizer that in English. Consider the following, where (a) is thatless, and (b)–(d) contain an explicit complementizer:

5.	a. SEAN:	.. Oh I think it would be hard (SBC: 051).
5.	b. JILL:	You mean you actually thought that we had seen this before?
		(SBC: 028).
5.	c. BEN:	.. (H) This assures us that n- any excess water coming down the
		Colorado River... (SBC: 038)
5.	d. LAJUAN:	(H) he calls me and tells me that he wants to be with me (SBC: 044).

According to Thompson and Mulac (1991b), it is not at all surprising that 5(a) is thatless, whereas (b), (c) and (d) are not. Thatless constructions are preferred with main clauses which have a subject I (or you, if the sentence is a question), a main verb present tense think (also guess), no indirect object, and no adverbial. The complement preferably has a pronominal subject. All these conditions are met by (a). Note that (b) has you for a non-question matrix subject, it includes an adverbial, and the verb is in the past tense. (c) has an inanimate matrix subject, a non-think verb, which takes an indirect object, and the complement subject is a lexical NP. In (d) the matrix subject is third person, the verb is neither think nor guess, and a direct object is included. No wonder (b), (c) and (d) each contain that. Thompson and Mulac support their claims with statistically significant data.
But why should discourse reflect any consistent patterns regarding +/– that, and why specifically the above tendencies? To answer this question we first need to consider what it is that speakers use complement-taking verb constructions for in natural discourse. Thompson’s (2002) proposal is that these constructions typically convey a speaker’s stance toward some proposition. The matrix part mostly expresses an epistemic evaluation, that is, how likely the embedded pro- position is to be true. As such, the matrix mainly serves as a subjective framing for the contextually more relevant information contained in the embedded clause. For example, in 6(a), the response to Sean’s thatless 5(a), inquires about
10 Discourse Studies 11(1)
the complement, what exactly would be hard, not about Sean’s thinking, but the response to the thatful 5(b), 6(b), does address the epistemic stance expressed in the matrix:
6. a. SEAN: .. Oh I think it would be hard. ALICE: ... What,
signing? (SBC: 051)

6. 
   b. JILL: You mean you actually thought that we had seen this before? JEFF: ... (H) No.
   

.
Actually tha- I I –
.. No I didn’t,
.. I just –
(H) I just.. thought that it was like pretty much factual (SBC: 028).
Once we realize what the frequent discourse function of these constructions is (the pattern in 5/6a, rather than that of 5b,c,d), the list of linguistic peculiarities above no longer seems surprising.
We can mostly report our own epistemic stances (hence the recurrent first person), although we can ask about the addressee’s stance (hence, second person in questions). The specific set of verbs which occur in the construction are the ones suitable for expressing stances. The pronominal subject of the com- plement testifies that it hooks up with the preceding context, quite often, it’s the topic of the utterance as a whole. Now, why no adverbials and no indirect objects? Because these would make the matrix too contentful and less of a mere epistemic framing for the complement. Finally, and most crucially, why is it that the more clearly epistemic framing cases preferably occur without a that? The complementizer serves as a clear signal that a separate, and specifically sub- ordinate clause follows. But if Thompson’s (2002) analysis is correct, then the so-called embedded clause is not interactionally subordinate at all, and the so- called matrix clause is not so much of an independent sentence, as it is a stance marker.⁴ As such, it’s closer to a sentential adverbial. If discoursally speaking, speakers analyze such constructions as an epistemic adverbial followed by a contextually relevant proposition which it modifies, we can understand why the discourse patterns created by these constructions are the way they are. Although grammatically speaking any subject, any adverbial and any indirect object can co-occur with a complement taking verb, speakers find only a restricted subset of these combinations to be useful for their discourse purposes. Clearly then, while grammar proposes a variety of options (all the above combinations), dis- coursal use very often calls for a careful and skewed selection of those options, tailoring the grammatical choice to the specific discourse context. This is what is meant by grammar proposing and discourse disposing.
Our next two examples are somewhat different. We see here that discourse patterns are skewed, not just because the messages we have use for in social interactions naturally call for some, but not other, grammatical patterns (as is the case for +/– that distribution). Rather, there are discourse-specific principles which directly dictate that we prefer some grammatical options over others.
Ariel: Discourse, grammar, discourse 11

2. 
   THE PREFERRED ARGUMENT STRUCTURE DISCOURSE PATTERN
   

The first example we consider concerns discourse preferences regarding argu- ment structures. A verb’s argument structure is the specification of the arguments (i.e. nominal roles, such as subject, direct object, indirect object) it requires or allows. Consider the following, all perfectly grammatical English sentences containing the verb meet:

6. 
   a. They never met until a long time down the road (LSAC)
   

7. b. Mark and I met at Alabama (LSAC)
7. c. BETH: ... I think I’d met her once before that though (SBC: 031)
7. d. WOOD: I met all the great artists of that day, (SBC: 05)
7. e. My mom’s met them (LSAC).

7. 
   f. LAJUAN: and my brother had met ~Ron, (SBC: 044).